AMD: Pirate Islands (R* 3** series) Speculation/Rumor Thread

[Razor1]

http://videocardz.com/56728/amd-radeon-r9-fury-x-reviewers-guide

review guide

 

[3dilettante]

http://www.legitreviews.com/amd-fiji-arrives-radeon-r9-fury-x-details_166515#OpQZsfoAmLt9ajjt.99



So the chip won't be bandwidth bottlenecked? That should be interesting.

It can be managed, at least if code tries hard enough. It sounds like teething pains, and at least temporarily takes away one of the benefits of HBM, per AMD.

AMD also claims overclocking should be much easier than with GDDR5, thanks to the simpler clock system, which requires less voltage. Notably, even a small increase in clock speed will result in a large increase in bandwidth, thanks to the wide bus, so it'll be interesting to see how far overclockers can push the technology.

http://arstechnica.com/information-...hbm-why-amds-high-bandwidth-memory-matters/2/

 

[Ryan Smith]

Anandtech confirms Fury has 1/16th FP64 compared to 1/8th in R9 290X/390X.

This time it might be a hardware limitation and not just pure software cap.

Correct. That's a hardware limitation. They hit the reticle limit and they weren't going to be able to put in over 4GB of VRAM. It was never going to be a FP64 compute GPU.

I'm not sure why Anandtech doesn't have typical board power numbers for the 290 series.
http://www.legitreviews.com/amd-radeon-r9-295x2-8gb-video-card-review-at-4k-ultra-hd_138950

The new cards use slightly more.

Originally AMD never published it. When I was writing that table I forgot that we finally got them to cough up a number when the 295X2 was launched (and have since corrected the article). Even then that's the only place you'll find it.

[wikipedia on UVD]

https://en.wikipedia.org/wiki/Unified_Video_Decoder

As usual Wikipedia is wrong. Officially Tonga doesn't have an HEVC decoder, though I wouldn't be surprised if it was maybe the same UVD design with HEVC either broken or intentionally turned off. Fiji however does have an HEVC decoder.

 

[lanek]

http://cdn.videocardz.com/1/2015/06/AMD-Radeon-R9-Fury-X-Overclock-performance.png

more benches

10% overclock on the core is giving ~5% performance increase according to this, so memory overclocking please.....

Seems completely differ from this one

Your posted one :

 

[pharma]

Ahh ... this is a fantastic disinformation campaign. Why not just release NDA if the Fury series is that good? We will find out about it sooner or later ...

 

[3dilettante]

It's only a suggestion that it's not the task of a new GPU to accelerate the stuff from yesteryear.

The big differentiator so far is HBM. It's a good direction going forward, but it's still just DRAM. It provides binary bits that taste very similar to GDDR5 bits.

Originally AMD never published it. When I was writing that table I forgot that we finally got them to cough up a number when the 295X2 was launched (and have since corrected the article). Even then that's the only place you'll find it.

I think that the omission was understandable if AMD could only be made to disclose it once over the course of multiple years.

 

[lanek]

Ahh ... this is a fantastic disinformation campaign. Why not just release NDA if the Fury series is that good? We will find out about it sooner or later ...

The question, is the first slide is coming from AMD, the second from.. well i dont know ..

 

[pharma]

The question, is the first slide is coming from AMD, the second from.. well i dont know ..

At this point they could both be from AMD ... different settings for same games (ie, Farcry 4).

 

[lanek]

At this point they could both be from AMD ... different settings for same games (ie, Farcry 4).

Yeah i have see that..

 

[Esrever]

Why would AMD even make the second chart is it shows them losing this badly?

 

[Ethatron]

The big differentiator so far is HBM. It's a good direction going forward, but it's still just DRAM. It provides binary bits that taste very similar to GDDR5 bits.

I don't really need differentiation, I need more of what I already got with GCN. Next iteration a bit more GPRs please.

 

[3dilettante]

I don't really need differentiation, I need more of what I already got with GCN. Next iteration a bit more GPRs please.

Is there an expectation that Fiji will have a broader feature set than Hawaii, a GPU of several years ago?
How long into the future can a GPU be excused for not accelerating games of yesteryear?

 

[SimBy]

Seems completely differ from this one
Your posted one :

This is an OC chart that has an error. Left and right are both Fury X. Right one is 100MHz OC.

 

[Razor1]

Yeah just saw that, the second table is from the reviewers guide and has an error the title of the 980ti

 

[SimBy]

So it seems Fiji does not really scale all that well when overclocking.

 

[lanek]

This is an OC chart that has an error. Left and right are both Fury X. Right one is 100MHz OC.

This explain all .. thank you...

 

[3dilettante]

At least there is some order to the universe.
Confirmed: a faster Fury is faster than Fury.

 

[Razor1]

Interesting though now with far cry 4 turning on x2 msaa the fury x looses 20% that is a huge hit.

Is Far Cry 4 one of the games that is close to the 4gb limit?

 

[SimBy]

Interesting though now with far cry 4 turning on x2 msaa the fury x looses 20% that is a huge hit.

Is Far Cry 4 one of the games that is close to the 4gb limit?

Yes I think it's right up there at 4GB @4K

http://www.hardocp.com/article/2015...cs_features_performance_review/7#.VYNePfmqqt4

 

[Razor1]

So it seems Fiji does not really scale all that well when overclocking.

Yeah it seems it needs more bandwidth also when overclocking the GPU